lecture note: department of physics, kitasato university november 15… · 2017. 11. 13. ·...
Post on 17-Mar-2021
1 Views
Preview:
TRANSCRIPT
Lecture Note:
Department of Physics, Kitasato University
November 15,16 2017:
Simplest Quantum Mechanics
Ryu Sasaki
Faculty of Science, Shinshu University,
Matsumoto 390-8621, Japan
Abstract
By pursuing the similarity and parallelism between the ordinary Quantum Mechan-ics (QM) and the eigenvalue problem of hermitian matrices, we present simplest formsof exactly solvable QM. Their Hamiltonians are a special class of tri-diagonal real sym-metric (Jacobi) matrices. The super (sub) diagonal elements of the Jacobi matricesare interpreted as the positive (negative) shift operators acting on vectors which areexpressed as functions defined on a finite (infinite) integer lattice, x = 0, 1, 2 . . .. Nowthe Jacobi matrices are second order difference operators which are Hamiltonians ofdiscrete Quantum Mechanics with real shifts (rdQM) in one dimension. Their eigen-functions (vectors) are the well known classical orthogonal polynomials of a discretevariable belonging to the Askey scheme of hypergeometric orthogonal polynomials, e.g.the Charlier, Meixner, (dual) Hahn, Racah etc and their q-versions. Here we presentthe simplest QM counterparts of the prominent results of exactly solvable QM; Crum’stheorem, shape invariance, Heisenberg operator solutions together with the dualityand dual polynomials which are the special features of the discrete QM with real shifts.Among many applications of the orthogonal polynomials of a discrete variable, we dis-cuss Birth and Death Processes, which are typical stationary Markov processes with1-d discrete state spaces.
PACS: 03.65.-w, 03.65.Ca, 03.65.Fd, 03.65.Ge, 02.30.Ik, 02.30.Gp, 02.70.Bf
keywords: matrix quantum mechanics, discrete quantum mechanics with real shifts,
exact solvability, factorised Hamiltonian, intertwining relations, shape invariance, Heisenberg
operator solutions, orthogonal polynomials of a discrete variable, duality, dual polynomials
Contents
1 Introduction 3
2 General Formulation 4
2.1 Problem Setting:1-d QM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Factorised Hamiltonian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2.1 Explicit Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2.2 Other Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.3 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3 Dual Polynomials 16
4 Fundamental Structure of Solution Space 20
5 Shape Invariance: Sufficient Condition of Exact Solvability 23
5.1 Shape Invariance Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
5.1.1 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
5.2 Auxiliary Function φ(x) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
5.2.1 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
5.3 Universal Rodrigues Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
5.3.1 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
6 Solvability in Heisenberg Picture 27
6.1 Closure Relation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
6.2 Creation and Annihilation Operators . . . . . . . . . . . . . . . . . . . . . . 30
6.3 Determination of An and Cn . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
6.4 Closure Relation Data and An, Cn . . . . . . . . . . . . . . . . . . . . . . . . 32
6.4.1 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
7 Applications: Birth and Death Processes 35
7.1 Schrodinger eqs. and Fokker-Planck eqs. . . . . . . . . . . . . . . . . . . . . 36
7.2 Birth & Death Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
8 Summary and Comments 38
2
A Symbols, Definitions & Formulas 39
B Three Term Recurrence Relations 40
1 Introduction
It is almost hundred years since the construction of quantum mechanics (QM). Together
with relativity, quantum theory constitutes the backbone of modern physics. It is very
useful and important for younger theoretical/mathematical physicists to acquire in depth
training of quantum mechanical thinking. This series of lectures is a modest attempt to
offer such an opportunity. Among many problem settings of QM, the eigenvalue problems,
which determine the energy spectrum and the corresponding eigenfunctions, are central.
Various techniques, e.g. exact solution methods [1, 2, 3] and different types of approximation
procedures, have been developed for over 90 years.
In this lecture, we revisit several known exact solution methods of one-dimensional QM
indirectly, by picking up a very closely related but much simpler problem; the eigenvalue
problem of a certain type of hermitian matrices, the simplest QM. We do hope that the
simple structure of the problem would make it easy to understand deeply the corresponding
quantum mechanical results. Moreover, as will become clear gradually, the present viewpoint
also provides a unified theory of orthogonal polynomials of a discrete variable belonging to
the Askey scheme of hypergeometric orthogonal polynomials [4, 5, 6]. The simplest QM
has now established itself as a branch of exactly solvable QM, called discrete QM with real
shifts (rdQM) [7, 8] with many new discoveries [9] and applications [10]. For an introductory
lecture on the exactly solvable ordinary QM, see [3].
This lecture note is organised as follows. In section 2 the tri-diagonal real symmetric
matrices are chosen as positive semi-definite Hamiltonians by following the same arguments
as the 1-d QM. The corresponding matrix eigenvalue problems are reinterpreted as second
order difference equations in a factorised form. Various examples, corresponding to the
Krawtchouk, Meixner, Charlier and (q-)Racah polynomials etc [6], are given by specifying the
potential functions. In section 3 the duality and dual polynomials are introduced by solving
the matrix equations through three term recurrence relations [7]. In section 4 the structure
of the solution spaces is explored through the intertwining relations (Crum’s theorem) [11],
which is a direct consequence of the factorised Hamiltonians. A sufficient condition for
3
exact solvability, the shape invariance, is explained in section 5. The universal eigenvalue
formula and Rodrigues formula are derived as direct consequences of the shape invariance.
In section 6, it is shown that all the examples listed in section 2 are exactly solvable in
Heisenberg picture, too [7, 12]. Through the closure relation, the creation/annihilation
operators are introduced. The coefficients of the three term recurrence relations, An and Cn
of the eigenpolynomials are determined [7]. The final section is for a summary and comments.
Appendix A provides a list of symbols, definitions and formulas. A short explanation of the
three term recurrence relation is given in Appendix B.
2 General Formulation
When we learn QuantumMechanics (QM), we notice close similarities between the eigenvalue
problems in QM and those of hermitian matrices;
QM: Hψ(x) = Eψ(x), (2.1)
H : self-adjoint H = H† (f, g)def=
∫f ∗(x)g(x)dx, (f,Hg) = (Hf, g), (2.2)
Matrix: Hv = λv,∑j
Hi,jvj = λvi, (2.3)
H : hermitian Hi,j = H∗j,i, (u, v)
def=
∑j
u∗jvj, (f,Hg) = (Hf, g). (2.4)
Their eigenvalues are real and eigenfunctions (vectors) belonging to distinct eigenvalues are
orthogonal:
λ ∈ R, Hψ1(x) = λ1ψ1(x), Hψ2(x) = λ2ψ2(x), λ1 = λ2, (ψ1, ψ2) = 0. (2.5)
2.1 Problem Setting:1-d QM
QM is based upon a very special type of self-adjoint operators. For simplicity, let us focus
on the simplest case of one dimensional QM. The Hamiltonian is a second order differential
operator corresponding to Newtonian mechanics:
H = − d2
dx2+ V (x), V (x) ∈ R. (2.6)
Many exactly solvable potentials and their eigenvalues, eigenfunctions are known [1, 2, 3].
The following three exactly solvable examples define the classical orthogonal polynomials,
4
the Hermite, Laguerre and Jacobi:
V (x) = x2 − 1, −∞ < x <∞, E(n) = 2n, ϕn(x) = e−x2/2Hn(x), n = 0, 1, 2, . . . , (2.7)
Hn(x) : Hermite Polynomial, (A.8),
V (x) = x2 +g(g − 1)
x2− (1 + 2g), 0 < x <∞, g > 1,
E(n) = 4n, ϕn(x) = e−x2/2xgL(g−1/2)
n (η(x)), n ∈ Z≥0, (2.8)
L(α)n (η) : Laguerre Polynomial, (A.9), η(x)
def= x2,
V (x) =g(g − 1)
sin2 x+h(h− 1)
cos2 x− (g + h)2, 0 < x < π/2, g, h > 1,
E(n) = 4n(n+ g + h), ϕn(x) = (sinx)g(cosx)hP (g−1/2,h−1/2)n (η(x)), n ∈ Z≥0, (2.9)
P (α,β)n (η) : Jacobi Polynomial, (A.10), η(x)
def= cos 2x. (2.10)
For these polynomials the square of the groundstate eigenfunction ϕ20(x), having no zero,
provides the orthogonality weight function
(ϕn, ϕm) = ((Pn, Pm))def=
∫ϕ20(x)Pn(η(x))Pm(η(x))dx = hnδnm, hn > 0. (2.11)
At this point very naive and natural questions arise:
Q1: Is there a special class of hermitian matrices corresponding to the ordinary
one-dimensional QM (2.6)?
Q2: Are there many exactly solvable ones whose eigenfunctions (vectors) define
orthogonal polynomials?
The answer to both questions is Yes. For Q1, it is called Jacobi matrices, which are tri-
diagonal and real symmetric matrices of finite or infinite dimensions. For Q2, they are
classical orthogonal polynomials of a discrete variable [13] whose orthogonality measures
consist of discrete point sets. They belong to the Askey scheme of hypergeometric orthogonal
polynomials [4, 6]. Those polynomials in the Askey scheme having the Jackson integral
measures, e.g. the big q-Jacobi, etc, are not included as they need a separate setting [14].
In the rest of this note we will explore the Simplest Quantum Mechanics.
2.2 Factorised Hamiltonian
In order to pursue the similarity with the one-dimensional QM, let us denote the indices of
the matrices by x, y = 0, 1, . . . xmax and the vector component by vx or v(x). The matrix
5
eigenvalue equation (2.3) is rewritten as
xmax∑y=0
Hx,yvy = λvx, or Hv(x) = λv(x),xmax∑y=0
Hx,yv(y) = λv(x),
in which xmax = N , a positive integer, for finite dimensional matrices and xmax = ∞ for
infinite matrices. For finite matrices, there are N + 1 eigenvalues. For infinite matrices we
restrict to those matrices having discrete eigenvalues only and their spectrum is bounded
from below. In other words, there is a lowest eigenvalue. Thus we denote the eigenvalue
problem as
Hϕn(x) = E(n)ϕn(x),∑y=0
Hx,yϕn(y) = E(n)ϕn(x), n = 0, 1, 2, . . . , (2.12)
in which the eigenvalues are numbered in the increasing order. We can always make the lowest
eigenvalue to be vanishing by subtracting E(0)× Identity Matrix from the Hamiltonian H:
0 = E(0) ≤ E(1) ≤ E(2) ≤ · · · .
Now the Hamiltonian, an hermitian matrix, is positive semi-definite. A linear algebra theo-
rem tells that a positive semi-definite hermitian matrix can always be expressed in a factorised
form:
H = A†A, 0 = det(A) = det(A†) = det(H), (2.13)
in which matrix A can be multiplied by any unitary matrix U , A → A′ = UA.
This is also the case for the positive semi-definite (E(0) = 0) Hamiltonian H in the
ordinary QM in one dimension (2.6). The groundstate eigenfunction ϕ0(x) satisfies
−d2ϕ0(x)
dx2+ V (x)ϕ0(x) = 0,
and it has no zero. Thus we obtain a non-singular expression of the potential function V (x)
V (x) =∂2xϕ0(x)
ϕ0(x)= ∂x
(∂xϕ0(x)
ϕ0(x)
)+
(∂xϕ0(x)
ϕ0(x)
)2
, (2.14)
and a factorised Hamiltonian:
H = A†A = − d2
dx2+ V (x), Aϕ0(x) = 0 ⇒ Hϕ0(x) = 0, (2.15)
A def=
d
dx− ∂xϕ0(x)
ϕ0(x), A† = − d
dx− ∂xϕ0(x)
ϕ0(x). (2.16)
6
We expect that the Hamiltonian H (2.13) is a second order difference operator consisting
of a positive unit shift ψ(x) → ψ(x+1) and the negative unit shift ψ(x) → ψ(x−1) operators.
For a continuous function f(x), Taylor expansion reads
f(x+ a) =(ea∂xf
)(x) =
∞∑n=0
an
n!∂nxf(x).
This suggests the following simplified notation for the unit shift operators
e±∂ ≡ e±∂x ;(e±∂f
)(x) = f(x± 1), (e∂)† = e−∂, (2.17)
which will be abused for the present case of discrete variables x = 0, 1, 2, . . .,
f(x± 1) =(e±∂f
)(x) =
∑y
(e±∂)x,yf(y) ⇒ (e±∂)x,y = δx±1,y.
As a matrix the positive (negative) unit shift operator e∂ (e−∂) is a super (sub) diagonal
matrix.
The factorisation operators A, A† in the ordinary 1-d QM case are a function plus a first
order differential operator. This suggests that A, A† in matrix QM (2.13) be a function plus
a first order difference operator e±∂, either positive e∂ or negative e−∂ shift operator, but not
both. Otherwise, the Hamiltonian would be a fourth order difference operator containing
e±2∂ as well as e±∂.
A simple guesswork would lead to the form
A =√B(x)− e∂
√D(x), A† =
√B(x)−
√D(x) e−∂, B(x), D(x) > 0, (2.18)
Ax,y =√B(x) δx,y −
√D(x+ 1) δx+1,y, (A†)x,y =
√B(x) δx,y −
√D(x) δx−1,y, (2.19)
with two positive potential functions B(x) and D(x). Here and hereafter we suppress the
identity matrix 1 = (δx,y),√B(x)1 →
√B(x). The resulting Hamiltonian is
H = B(x) +D(x)−√B(x) e∂
√D(x)−
√D(x) e−∂
√B(x), (2.20)
Hx,y =(B(x) +D(x)
)δx,y −
√B(x)D(x+ 1) δx+1,y −
√B(x− 1)D(x) δx−1,y. (2.21)
In order that (Hψ) (x) does not contain undefined quantities ψ(−1) or ψ(N + 1) for the
finite case, the positive potential functions must vanish at the boundary
B(x), D(x) > 0, D(0) = 0, B(N) = 0. (2.22)
7
As seen clearly from the two types of equivalent expressions for the Hamiltonian H (2.20),
(2.21) or A, A† (2.18), (2.19), one can treat them either as matrices (2.21), (2.19) or as
difference operators (2.20), (2.18) acting on functions. For actual calculations, however, the
difference operators are much easier to handle than explicit matrix forms.
Thus the Hamiltonian H is a tri-diagonal real symmetric matrix:
H =
B(0) −√B(0)D(1) 0 · · · · · · 0
−√B(0)D(1) B(1) +D(1) −
√B(1)D(2) 0 · · ·
..
.
0 −√B(1)D(2) B(2) +D(2) −
√B(2)D(3) · · ·
..
.... · · · · · · · · · · · ·
.
..... · · · · · · · · · · · · 0
0 · · · · · · −√B(N−2)D(N−1) B(N−1)+D(N−1) −
√B(N−1)D(N)
0 · · · · · · 0 −√B(N−1)D(N) D(N)
.
This type of matrices are called Jacobi matrices. It is known that all the eigenvalues are
simple:
0 = E(0) < E(1) < E(2) < · · · < E(N) (< · · · ). (2.23)
This guarantees the orthogonality of all eigenvectors. The factorisation matrices A and A†
are
A =
√B(0) −
√D(1) 0 · · · · · · 0
0√B(1) −
√D(2) 0 · · · ...
0 0√B(2) −
√D(3) · · · ...
... · · · · · · · · · · · · ...
... · · · · · · · · · · · · 0
0 · · · · · · 0√B(N − 1) −
√D(N)
0 · · · · · · 0 0 0
, (2.24)
A† =
√B(0) 0 0 · · · · · · 0
−√D(1)
√B(1) 0 0 · · · ...
0 −√D(2)
√B(2) 0 · · · ...
... · · · · · · · · · · · · ...
... · · · · · · · · · · · · 0
0 · · · · · · −√D(N − 1)
√B(N − 1) 0
0 · · · · · · 0 −√D(N) 0
. (2.25)
From these it is obvious 0 = det(A) = det(A†) = det(H) for the finite case. For the infinite
dimensional case,∏∞
x=0B(x) = 0 is needed.
8
The ground state eigenfunction (vector) ϕ0(x), Hϕ0(x) = 0, is easily obtained as the zero
mode of the operator A, Aϕ0(x) = 0, as in the ordinary QM (2.16):
Aϕ0(x) = 0 ⇒√B(x)ϕ0(x)−
√D(x+ 1)ϕ0(x+ 1) = 0, (2.26)
⇒ ϕ0(x) =x−1∏y=0
√B(y)
D(y + 1), x = 1, 2, . . . , (2.27)
with the boundary condition
ϕ0(0) = 1. (2.28)
As in the ordinary QM, the groundstate wavefunction has no zero.
We look for the solution of the matrix eigenvalue problem (2.12) by assuming the fac-
torised form
ϕn(x) = ϕ0(x)Pn(x), (2.29)
as in the exactly solvable examples in ordinary QM (2.7)–(2.9). In the simplest QM, one can
interpret Pn(x) either as a function or a vector with the x component Pn(x). In the latter
interpretation, ϕ0(x) in (2.29) is understood as a diagonal matrix. The matrix eigenvalue
problem Hϕn(x) = E(n)ϕn(x) (2.12) is now rewritten as
(B(x) +D(x))ϕ0(x)Pn(x)−√B(x)D(x+ 1)ϕ0(x+ 1)Pn(x+ 1)
−√B(x− 1)D(x)ϕ0(x− 1)Pn(x− 1) = E(n)ϕ0(x)Pn(x).
By using the zero mode equation for ϕ0(x) (2.26) and dividing by ϕ0(x), we arrive at a simple
difference equation for Pn(x):
B(x)(Pn(x)− Pn(x+ 1)
)+D(x)
(Pn(x)− Pn(x− 1)
)= E(n)Pn(x), (2.30)
⇒ HPn(x) = E(n)Pn(x), H1 = 0, (2.31)
H def= ϕ0(x)
−1 · H · ϕ0(x) = B(x)(1− e∂) +D(x)(1− e−∂). (2.32)
This result justifies the rather oddly looking√B(x) and
√D(x) in the definitions of A
and A† (2.18). For the matrix interpretation of H (2.32), ϕ0(x) is a diagonal matrix,
ϕ0(x) = diag{ϕ0(0), ϕ0(1), · · · , ϕ0(N)}, instead of a groundstate eigenvector. For the dif-
ference operator interpretation of H (2.32), ϕ0(x) is a function and it would be suitable to
rewrite (2.32)
H def= ϕ0(x)
−1 ◦ H ◦ ϕ0(x), (2.33)
9
as in the ordinary QM.
A simple sufficient condition for the solvability of the similarity transformed Hamiltonian
H is the triangularity with respect to a special basis
1, η(x), η(x)2, . . . , η(x)n, . . . , (2.34)
spanned by a certain function η(x) which is called sinusoidal coordinate [12]:
Hη(x)n = E(n)η(x)n + lower orders in η(x). (2.35)
This property is preserved under the affine transformation of η(x), η(x) → aη(x) + b (a, b:
constants). We choose the boundary condition
η(0) = 0. (2.36)
The triangularity (2.35) implies that the eigenvectors can be obtained as polynomials in η(x)
(Pn(x) ≡ Pn(η(x))) by finite linear algebra up to normalisation:
HPn(η(x)) = E(n)Pn(η(x)),
Hϕn(x) = E(n)ϕn(x), ϕn(x) = ϕ0(x)Pn(η(x)),n = 0, 1, 2, . . . , (N) , . . . . (2.37)
We choose a simple universal normalisation of Pn(η(x))
Pn(0) = 1 ⇔ Pn(0) = 1, n = 0, 1, 2, . . . , (N) , . . . , (2.38)
which also means
P0 = 1. (2.39)
The eigenpolynomials {Pn(η(x))} can also be obtained by the Gram-Schmidt orthogonalisa-
tion of the sinusoidal coordinates (2.34) with the weight function ϕ20(x).
Here are five known forms of the sinusoidal coordinates in rdQM [7] (0 < q < 1):
(i) η(x) = x, (ii) η(x) = ϵ′x(x+ d), (ϵ′ = 1 for d > −1, ϵ′ = −1 for d < −N)
(iii) η(x) = 1− qx, (iv) η(x) = q−x − 1,
(v) η(x) = ϵ′(q−x − 1)(1− dqx), (ϵ′ = 1 for d < q−1, ϵ′ = −1 for d > q−N).
(2.40)
2.2.1 Explicit Examples
Here are some examples of simplest QM [7], which are called by the name of the eigenpoly-
nomials [6]. One finite dimensional example is the Krawtchouk and two from infinite ones
are the Meixner and the Charlier.
10
Krawtchouk The potential functions are
B(x) = p(N − x), D(x) = (1− p)x, 0 < p < 1, η(x) = x. (2.41)
The triangularity reads
Hxn = p(N − x)(xn − (x+ 1)n
)+ (1− p)x
(xn − (x− 1)n
)= nxn + lower degrees, E(n) = n.
The eigenpolynomials are (see (A.3))
Pn(x; p,N) = 2F1
(−n, −x−N
∣∣∣ 1p
), P0 = 1, P1 =
pN − x
pN,
P2 =p2N(N − 1) + (2p(1−N)− 1)x+ x2
p2N(N − 1), . . . .
(2.42)
The orthogonality weight function is
ϕ20(x) =
x−1∏y=0
p
1− p
N − y
y + 1=
(p
1− p
)xN !
x!(N − x)!=
(p
1− p
)xN !
Γ(x+ 1)Γ(N + 1− x). (2.43)
The trace formula is easy to verify:
N
2(N + 1) =
N∑n=0
E(n) = Tr(H) =N∑x=0
(B(x) +D(x)) . (2.44)
Meixner The potential functions are
B(x) =c
1− c(x+ β), D(x) =
x
1− c, β > 0, 0 < c < 1, η(x) = x. (2.45)
The triangularity reads
(1− c)Hxn = c(x+ β)(xn − (x+ 1)n
)+ x
(xn − (x− 1)n
)= (1− c)nxn + lower degrees, E(n) = n.
The eigenpolynomials are
Pn(x; β, c) = 2F1
(−n, −xβ
∣∣∣ 1− 1
c
), P0 = 1, P1 =
βc− (1− c)x
βc,
P2 =β(β + 1)c2 + (−1 + c)(1 + c+ 2βc)x+ (−1 + c)2x2
β(β + 1)c2, . . . .
(2.46)
11
The orthogonality weight function is square summable:
ϕ20(x) =
x−1∏y=0
c(y + β)
y + 1=
(β)xcx
x!=
Γ(β + x) cx
Γ(β)Γ(x+ 1),
∞∑x=0
ϕ20(x) =
∞∑x=0
(β)xcx
x!= 1F0
( β−
∣∣∣ c) =1
(1− c)β,
(2.47)
in which (a)n is the shifted factorial (Pochhammer symbol) (A.1).
Charlier The potential functions are
B(x) = a, D(x) = x, 0 < a, η(x) = x. (2.48)
The triangularity is trivial
Hxn = a(xn − (x+ 1)n
)+ x
(xn − (x− 1)n
)= nxn + lower degrees, E(n) = n.
The eigenpolynomials are
Pn(x; a) = 2F0
(−n, −x−
∣∣∣ −1
a
), P0 = 1, P1 =
a− x
a,
P2 =a2 − (1 + 2a)x+ x2
a2, . . . .
(2.49)
The orthogonality weight function is square summable:
ϕ20(x) =
x−1∏y=0
a
y + 1=ax
x!=
ax
Γ(x+ 1),
∞∑x=0
ϕ20(x) =
∞∑x=0
ax
x!= ea. (2.50)
In all these examples ϕ20(x) is a meromorphic function of x and it vanishes on the integer
points outside of the domain:
ϕ20(x) = 0, x ∈ Z\{0, 1, . . . , N}, or x ∈ Z<0. (2.51)
2.2.2 Other Examples
Here are some other examples [6] without specifying the parameter ranges. The triangularity
of the polynomial equations (2.35) is purely algebraic and independent of the parameter
ranges. As shown in the examples, B(x), D(x) are rational functions of x or q±x and the
polynomial solutions are well defined in the whole complex plane x ∈ C. The parameter
12
ranges are restricted by the positivity of the potential B(x), D(x), the sinusoidal coordinate
η(x) and the eigenvalues E(n). The positivity of the orthogonality weight function ϕ20(x) is
not guaranteed for the parameters outside of the ranges.
Finite dimensional examples are followed by infinite dimensional ones. Within each
group, the most generic one is followed by simpler examples. The q-polynomials come after
non q-polynomials. We use the naming of the orthogonal polynomials used in [6]. The
parametrisation of some polynomials (Racah, (dual) Hahn and their q-versions) are different
from the conventional ones. We advocate using the present parametrisations for their obvious
symmetries. For the shifted q-factorials (a; q)n, (a1, . . . , ar; q)n, etc., see Appendix A (A.2).
Finite dimensional examples.
Racah One of the parameters a, b, c is identified with −N .
B(x) = −(x+ a)(x+ b)(x+ c)(x+ d)
(2x+ d)(2x+ 1 + d), D(x) = −(x+ d− a)(x+ d− b)(x+ d− c)x
(2x− 1 + d)(2x+ d),
η(x) = x(x+ d), E(n) = n(n+ d), ddef= a+ b+ c− d− 1,
ϕ20(x) =
(a, b, c, d)x(1 + d− a, 1 + d− b, 1 + d− c, 1)x
2x+ d
d.
(2.52)
Hahn
B(x) = (x+ a)(N − x), D(x) = x(b+N − x), η(x) = x,
E(n) = n(n+ a+ b− 1), ϕ20(x) =
N !
x! (N − x)!
(a)x (b)N−x
(b)N.
(2.53)
dual Hahn
B(x) =(x+ a)(x+ a+ b− 1)(N − x)
(2x− 1 + a+ b)(2x+ a+ b), D(x) =
x(x+ b− 1)(x+ a+ b+N − 1)
(2x− 2 + a+ b)(2x− 1 + a+ b),
η(x) = x(x+ a+ b− 1), E(n) = n,
ϕ20(x) =
N !
x! (N − x)!
(a)x (2x+ a+ b− 1)(a+ b)N(b)x (x+ a+ b− 1)N+1
.
(2.54)
13
q-Racah One of the parameters a, b, c is identified with q−N .
B(x) = −(1− aqx)(1− bqx)(1− cqx)(1− dqx)
(1− dq2x)(1− dq2x+1),
D(x) = −d (1− a−1dqx)(1− b−1dqx)(1− c−1dqx)(1− qx)
(1− dq2x−1)(1− dq2x), d
def= abcd−1q−1,
η(x) = (q−x − 1)(1− dqx), E(n) = (q−n − 1)(1− dqn),
ϕ20(x) =
(a, b, c, d ; q)x
(a−1dq, b−1dq, c−1dq, q ; q)x dx1− dq2x
1− d.
(2.55)
q-Hahn
B(x) = (1− aqx)(qx−N − 1), D(x) = aq−1(1− qx)(qx−N − b),
η(x) = q−x − 1, E(n) = (q−n − 1)(1− abqn−1),
ϕ20(x) =
(q ; q)N(q ; q)x (q ; q)N−x
(a; q)x (b ; q)N−x
(b ; q)N ax.
(2.56)
dual q-Hahn
B(x) =(qx−N − 1)(1− aqx)(1− abqx−1)
(1− abq2x−1)(1− abq2x),
D(x) = aqx−N−1 (1− qx)(1− abqx+N−1)(1− bqx−1)
(1− abq2x−2)(1− abq2x−1),
η(x) = (q−x − 1)(1− abqx−1), E(n) = q−n − 1,
ϕ20(x) =
(q ; q)N(q ; q)x (q ; q)N−x
(a, abq−1 ; q)x(abqN , b ; q)x ax
1− abq2x−1
1− abq−1.
(2.57)
quantum q-Krawtchouk
B(x) = p−1qx(qx−N − 1), D(x) = (1− qx)(1− p−1qx−N−1),
η(x) = q−x − 1, E(n) = 1− qn, ϕ20(x) =
(q ; q)N(q ; q)x(q ; q)N−x
p−xqx(x−1−N)
(p−1q−N ; q)x.
(2.58)
q-Krawtchouk
B(x) = qx−N − 1, D(x) = p(1− qx), η(x) = q−x − 1,
E(n) = (q−n − 1)(1 + pqn), ϕ20(x) =
(q ; q)N(q ; q)x(q ; q)N−x
p−xq12x(x−1)−xN .
(2.59)
14
dual q-Krawtchouk
B(x) =(qx−N − 1)(1 + pqx)
(1 + pq2x)(1 + pq2x+1), D(x) = pq2x−N−1 (1− qx)(1 + pqx+N)
(1 + pq2x−1)(1 + pq2x),
η(x) = (q−x − 1)(1 + pqx), E(n) = q−n − 1,
ϕ20(x) =
(q ; q)N(q; q)x(q; q)N−x
(−p ; q)x(−pqN+1 ; q)x pxq
12x(x+1)
1 + pq2x
1 + p.
(2.60)
affine q-Krawtchouk
B(x) = (qx−N − 1)(1− pqx+1), D(x) = pqx−N(1− qx),
η(x) = q−x − 1, E(n) = q−n − 1, ϕ20(x) =
(q ; q)N(q ; q)x(q ; q)N−x
(pq ; q)x(pq)x
.(2.61)
Infinite dimensional examples.
little q-Jacobi
B(x) = a(q−x − bq), D(x) = q−x − 1, η(x) = 1− qx,
E(n) = (q−n − 1)(1− abqn+1), ϕ20(x) =
(bq ; q)x(q ; q)x
(aq)x.(2.62)
little q-Laguerre This is obtained by setting b = 0 of the little q-Jacobi (2.62).
B(x) = aq−x, D(x) = q−x − 1, η(x) = 1− qx, E(n) = q−n − 1, ϕ20(x) =
(aq)x
(q ; q)x.
(2.63)
Al-Salam-Carlitz II
B(x) = aq2x+1, D(x) = (1− qx)(1− aqx), η(x) = q−x − 1,
E(n) = 1− qn, ϕ20(x) =
axqx2
(q, aq ; q)x.
(2.64)
alternative q-Charlier
B(x) = a, D(x) = q−x − 1, η(x) = 1− qx,
E(n) = (q−n − 1)(1 + aqn), ϕ20(x) =
axqx(x+1)/2
(q ; q)x.
(2.65)
The absence of the q-Meixner and q-Charlier
q-Meixner; B(x) = cqx(1− bqx+1), D(x) = (1− qx)(1 + bcqx),
15
η(x) = q−x − 1, E(n) = 1− qn, (2.66)
q-Charlier; B(x) = aqx, D(x) = 1− qx, η(x) = q−x − 1, E(n) = 1− qn, (2.67)
in the above list must be stressed, although their orthogonality weight functions in references
do not contain Jackson integrals. This is because the set of eigenvectors of the difference
Schrodinger equation (2.12) is not complete for these two polynomials [14].
2.2.3 Exercise
1. Verify the triangularity, E(n) and ϕ20(x) for the other examples in §2.2.2.
2. Calculate the lower members of the eigenpolynomials P1 and P2 explicitly for some
examples.
3. Verify the trace formula (2.44) for the finite dimensional examples.
4. Derive the meromorphic expressions of ϕ20(x) and confirm that they vanish outside of
the domain (2.51). For q-polynomials use the formula (a ; q)n = (a ; q)∞/(aqn ; q)∞
(A.2).
3 Dual Polynomials
The orthogonality of the eigenvectors of the Jacobi matrix H (2.20), (2.21) implies∑x
ϕ20(x)Pn(η(x))Pm(η(x)) =
1
d2nδn,m, (3.1)
in which we choose the normalisation constants {dn > 0} to be in the denominator. In terms
of the normalised eigenvectors
ϕ0(x) = d0ϕ0(x), ϕn(x) = dnϕ0(x)Pn(η(x)) = ϕ0(x)dnd0Pn(η(x)),
the orthonormality relation reads∑x
dnϕ0(x)Pn(η(x)) · dmϕ0(x)Pm(η(x)) = δn,m.
For the complete set of eigenvectors this means∑n
dnϕ0(x)Pn(η(x)) · dnϕ0(y)Pn(η(y)) = δx,y,
16
which states the simplest duality that the row eigenvectors are also orthogonal. When written
slightly differently ∑n
d2nPn(η(x))Pn(η(y)) =1
ϕ20(x)
δx,y, (3.2)
it displays the duality with (3.1) explicitly.
As Pn(η) is an orthogonal polynomial, it satisfies three term recurrence relations (see
Appendix B)
ηPn(η) = AnPn+1(η) +BnPn(η) + CnPn−1(η), n = 0, 1, . . . , (3.3)
with real coefficients An, Bn and Cn (C0 = 0). With the relation
Bn = −(An + Cn), (3.4)
which is a consequence of the universal normalisation (boundary) condition Pn(0) = 1 (2.38),
the above three term recurrence relation takes the same form as the H equation (2.30) in
the n-variable:
− An(Pn(η)− Pn+1(η)
)− Cn
(Pn(η)− Pn−1(η)
)= ηPn(η),[
−An(1− e∂n)− Cn(1− e−∂n)]Pn(η) = ηPn(η),
(3.5)
in which η is the eigenvalue.
The dual polynomial is an important concept in the theory of orthogonal polynomials of
a discrete variable [4, 5, 6, 13, 15, 16]. They arise naturally as the solution of the original
eigenvalue problem (2.12) or (2.30) obtained in a different way. Let us rewrite the similarity
transformed eigenvalue problem Hψ(x) = E ψ(x) (2.31) into an explicit matrix form with
the change of the notation ψ(x) → t(Q0, Q1, . . . , Qx, . . .)∑y
Hx,yQy = EQx, x, y = 0, 1, . . . , (N), . . . . (3.6)
Because of the tri-diagonality of H, it is in fact a three term recurrence relation for Qx as a
polynomial in E :
EQx(E) = B(x)(Qx(E)−Qx+1(E)
)+D(x)
(Qx(E)−Qx−1(E)
),
x = 0, 1, . . . , (N), . . . . (3.7)
17
Starting with the boundary (initial) condition
Q0 = 1, (3.8)
we determine Qx(E) as a degree x polynomial in E . It is easy to see that Qx(E) also satisfies
the universal normalisation condition (2.38),
Qx(0) = 1, x = 0, 1, . . . , (N), . . . . (3.9)
Replacing E by the actual value of the n-th eigenvalue E(n) (2.35) in Qx(E), we obtain the
explicit form of the eigenfunction (vector)∑y
Hx,yQy(E(n)) = E(n)Qx(E(n)), x = 0, 1, . . . , (N), . . . . (3.10)
In the finite dimensional case, {Q0(E), . . . , Qx(E), . . . , QN(E)} as degree x polynomials in Eare determined by (3.7) for x = 0, . . . , N − 1. The last equation
EQN(E) = D(N)(QN(E)−QN−1(E)
), (3.11)
is a degree N + 1 algebraic equation (characteristic equation) of E for the determination of
all the eigenvalues {E(n)}.Here are five known forms of the eigenvalues in rdQM, which are essentially of the same
forms as the sinusoidal coordinates (2.40) (0 < q < 1):
(i) : E(n) = n, (ii) : E(n) = ϵn(n+ α), (ϵ = 1 for α > −1,−1 for α < −N)
(iii) : E(n) = 1− qn, (iv) : E(n) = q−n − 1,
(v) : E(n) = ϵ(q−n − 1)(1− αqn), (ϵ = 1 for α < q−1, −1 for α > q−N).
(3.12)
Convince yourself that all the eigenvalues of the examples in §2.2.1, §2.2.2 are of these forms.
In subsequent sections we will provide further two independent algebraic methods of the
determination of the eigenvalues {E(n)} based on the shape-invariance (in §5) and the exact
Heisenberg operator solution in §6.We now have two expressions (polynomials) for the eigenvectors of the problem (2.30)
belonging to the eigenvalue E(n); Pn(η(x)) and Qx(E(n)). Due to the simplicity of the
spectrum of the Jacobi matrix, they must be equal up to a multiplicative factor αn,
Pn(η(x)) = αnQx(E(n)), x = 0, 1, . . . , (N), . . . , (3.13)
18
which turns out to be unity because of the boundary (initial) condition at x = 0 (2.36),
(2.38), (3.8) (or at n = 0 (2.28), (2.39), (3.9));
Pn(η(0)) = Pn(0) = 1 = Q0(E(n)), n = 0, 1, . . . , (N), . . . , (3.14)
P0(η(x)) = 1 = Qx(0) = Qx(E(0)), x = 0, 1, . . . , (N), . . . . (3.15)
We have established that the two polynomials, {Pn(η)} and its dual polynomial {Qx(E)},coincide at the integer lattice points:
Pn(η(x)) = Qx(E(n)), n = 0, 1, . . . , (N), . . . , x = 0, 1, . . . , (N), . . . . (3.16)
The completeness relation (3.2)∑n
d2nPn(η(x))Pn(η(y)) =∑n
d2nQx(E(n))Qy(E(n)) =1
ϕ20(x)
δx,y, (3.17)
is now the orthogonality relation of the dual polynomial Qx(E(n)), and the previous normal-
isation constant d2n is now the orthogonality weight function.
The real symmetric (hermitian) matrix Hx,y (2.21) can be expressed in terms of the
complete set of the eigenvalues and the corresponding normalised eigenvectors (spectral
representation)
Hx,y =∑n
E(n)ϕn(x)ϕn(y), (3.18)
ϕn(x) = dnϕ0(x)Pn(η(x)) = dnϕ0(x)Qx(E(n)). (3.19)
The very fact that it is tri-diagonal can be easily verified by using the difference equation
for the polynomial Pn(η(x)) (2.30) or the three term recurrence relation for Qx(E(n)) (3.7).Here is a list of the dual correspondence:
x↔ n, η(x) ↔ E(n), η(0) = 0 ↔ E(0) = 0,
B(x) ↔ −An, D(x) ↔ −Cn,ϕ0(x)
ϕ0(0)↔ dn
d0.
(3.20)
In the last expression, we have inserted ϕ0(0) = 1 (2.27) for symmetry. It should be remarked
that B(x) and D(x) govern the difference equation for the polynomial Pn(η), the solution
of which requires the knowledge of the sinusoidal coordinate η(x). The same quantities
B(x) and D(x) specify the three term recurrence of the dual polynomial Qx(E) without
the knowledge of the spectrum E(n). The explicit forms of η(x) and E(n) are required for
19
the polynomials Pn(η) and Qx(E) to be the eigenvectors of the eigenvalue problem (2.12).
Likewise, An and Cn in (3.3) specify the polynomial Pn(η) without the knowledge of the
sinusoidal coordinate. As for the dual polynomial Qx(E(n)), An and Cn provide the differenceequation (in n) (3.5), the solution of which needs the explicit form of E(n). Let us stress
that it is the eigenvalue problem (2.12) with the specific Hamiltonian (2.20) that determines
the polynomial Pn(η) and its dual Qx(E), the spectrum E(n), the sinusoidal coordinate η(x)and the orthogonality measures ϕ2
0(x) and d2n.
The three explicit examples in §2.2.1, the Krawtchouk (2.41), Meixner (2.45) and Char-
lier (2.48) are self-dual, as the explicit forms of the polynomials 2F1
(−n,−x−N
∣∣∣ 1p
)(2.42),
2F1
(−n,−xβ
∣∣∣ 1 − 1c
)(2.46), and 2F0
(−n,−x
−
∣∣∣ − 1a
)(2.49) are invariant under the exchange
x↔ n.
4 Fundamental Structure of Solution Space
Let us explore the structure of the solution space of the eigenvalue problem for the positive
semi-definite Jacobi matrices:
Hϕn(x) = E(n)ϕn(x), n = 0, 1, 2, . . . , (4.1)
H ≡ H[0] = A†A, H[0]ϕ0(x) = 0, Aϕ0(x) = 0, E(0) = 0. (4.2)
By changing the order of A† and A, a new hermitian matrix H[1] is defined:
H[1] def= AA†, H[0] = A†A, ϕ[0]
n (x) ≡ ϕn(x), n = 0, 1, . . . , . (4.3)
It is again a tri-diagonal symmetric matrix. If the original Hamiltonian H[0] is an (N +1)×(N +1) matrix, the new one H[1] is an N ×N matrix, as seen clearly from the explicit forms
of A (2.24) and A† (2.25). This is due to the boundary condition B(N) = 0. The new one
H[1] is called the partner or the associated Hamiltonian.
These two Hamiltonians are connected by the following intertwining relations :
AH[0] = AA†A = H[1]A, A†H[1] = A†AA† = H[0]A†. (4.4)
The iso-spectrality of the Hamiltonians H[0] and H[1] except for the groundstate ϕ[0]0 (x),
H[0]ϕ[0]n (x) = E(n)ϕ[0]
n (x) (n = 0, 1, . . .), Aϕ[0]0 (x) = 0, (4.5)
H[1]ϕ[1]n (x) = E(n)ϕ[1]
n (x) (n = 1, 2, . . .), (4.6)
20
ϕ[1]n (x)
def= Aϕ[0]
n (x), ϕ[0]n (x) =
A†
E(n)ϕ[1]n (x) (n = 1, 2, . . .), (4.7)
(ϕ[1]n , ϕ
[1]m ) = E(n)(ϕ[0]
n , ϕ[0]m ) (n,m = 1, 2, . . .). (4.8)
is a simple consequence of the intertwining relations. The groundstate energy of the partner
Hamiltonian H[1] is E(1) as the original ϕ0(x) is deleted, Aϕ0(x) = 0.
By subtracting the groundstate energy E(1) from the diagonal part of H[1], another
positive semi-definite Hamiltonian H[1] − E(1) is obtained, which is again factorised
H[1] = A[1]†A[1] + E(1), B[1](x), D[1](x) > 0, D[1](0) = 0, B[1](N − 1) = 0, (4.9)
A[1] =√B[1](x)− e∂
√D[1](x), A[1]† =
√B[1](x)−
√D[1](x) e−∂,
in the same manner as the original Hamiltonian
A[1]ϕ[1]1 (x) = 0. (4.10)
By changing the order of A[1]† and A[1], a new tri-diagonal Hamiltonian H[2] is defined:
H[2] def= A[1]A[1]† + E(1). (4.11)
The two Hamiltonians H[1] − E(1) and H[2] − E(1) are intertwined by A[1] and A[1]†:
A[1](H[1] − E(1)
)= A[1]A[1]†A[1] =
(H[2] − E(1)
)A[1],
A[1]†(H[2] − E(1))= A[1]†A[1]A[1]† =
(H[1] − E(1)
)A[1]†.
The iso-spectrality of H[1] and H[2] and the relations of their eigenvectors are obtained as
before:
H[2]ϕ[2]n (x) = E(n)ϕ[2]
n (x) (n = 2, 3, . . .),
ϕ[2]n (x)
def= A[1]ϕ[1]
n (x), ϕ[1]n (x) =
A[1]†
E(n)− E(1)ϕ[2]n (x) (n = 2, 3, . . .),
(ϕ[2]n , ϕ
[2]m ) =
(E(n)− E(1)
)(ϕ[1]
n , ϕ[1]m ) (n,m = 2, 3, . . .),
H[2] = A[2]†A[2] + E(2), A[2]ϕ[2]2 (x) = 0.
By the transformation H[1] → H[2], the groundstate corresponding to E(1) is deleted,
A[1]ϕ[1]1 (x) = 0.
This process of deleting the lowest energy state goes successively:
H[s] def= A[s−1]A[s−1]† + E(s− 1) = A[s]†A[s] + E(s), (4.12)
21
H[s]ϕ[s]n (x) = E(n)ϕ[s]
n (x) (n = s, s+ 1, . . .), A[s]ϕ[s]s (x) = 0, (4.13)
ϕ[s]n (x)
def= A[s−1]ϕ[s−1]
n (x), ϕ[s−1]n (x) =
A[s−1]†
E(n)− E(s− 1)ϕ[s]n (x)
(n = s, s+ 1, . . .), (4.14)
(ϕ[s]n , ϕ
[s]m ) =
(E(n)− E(s− 1)
)(ϕ[s−1]
n , ϕ[s−1]m ) (n,m = s, s+ 1, . . .). (4.15)
A
A†
A
A†
A
A†
A
A†
A[1]
A[1] †
A[1]
A[1] †
A[1]
A[1] †
A[2]
A[2] †
A[2]
A[2] †
A[3]
A[3] †
E(0)
E(1)
E(2)
E(3)
...
φ1
φ2
φ3
...
φ0
φ[1]1
φ[1]2
φ[1]3
φ[2]2
φ[2]3 φ
[3]3
. ..
H H[1] H[2] H[3] . . .
Figure 1: Fundamental structure of the solution space of 1-d QM.
Summarising these results we arrive at the following
Theorem 4.1 (Crum [11]) For a given Hamiltonian system {H[0], E(n), ϕ[0]n (x)} there ex-
ist as many associated Hamiltonian systems {H[1], E(n), ϕ[1]n (x)}, {H[2], E(n), ϕ[2]
n (x)}, . . . ,
as the number of discrete eigenvalues. They share the eigenvalues {E(n)} of the original
Hamiltonian and the eigenvectors of H[j] and H[j+1] are related by A[j] and A[j]†.
It should be stressed that Crum’s theorem is the consequence of the factorisation of
positive semi-definite Hamiltonians. It holds equally well in ordinary 1-d QM and in simplest
QM.
22
5 Shape Invariance: Sufficient Condition of Exact Solv-
ability
The structure of the solution space gives a key to understand the solvability. That is shape
invariance, which is a sufficient condition for exact solvability. In most cases Hamiltonians
depend on several parameters, λ = (λ1, λ2, . . .). Let us symbolically express the parameter
dependence of various quantities as H(λ), A(λ), E(n;λ), ϕn(x;λ), Pn(η(x);λ) etc.The system is exactly solvable if the matrix A[1] of the partner Hamiltonian H[1] has the
same form as A ≡ A[0] with a certain parameter shift:
A[1](λ) ∝ A[0](λ+ δ),
in which δ is the shift of the parameters. This can be rephrased as
A(λ)A(λ)† = κA(λ+ δ)†A(λ+ δ) + E(1;λ), (5.1)
in which κ > 0 is a parameter corresponding to the overall scale change of the Hamiltonians.
From Fig. 1, we find that the eigenvalues and the corresponding eigenvectors are expressed
by the first excited energy E(1;λ), A(λ), A(λ)† and the groundstate vector ϕ0(x;λ) at
variously shifted values of λ [7]:
E(n;λ) =n−1∑s=0
κsE(1;λ[s]), λ[s] def= λ+ sδ, (5.2)
ϕn(x;λ) ∝ A(λ[0])†A(λ[1])†A(λ[2])† · · · A(λ[n−1])†ϕ0(x;λ[n]). (5.3)
As we will see shortly, the eigenvector formula (5.3) is the generalised Rodrigues formula. In
terms of the potential functions the shape invariance condition (5.1) is rewritten as
B(x+ 1;λ)D(x+ 1;λ) = κ2B(x;λ+ δ)D(x+ 1;λ+ δ), (5.4)
B(x;λ)+D(x+ 1;λ) = κ(B(x;λ+ δ)+D(x;λ+ δ)
)+E(1;λ). (5.5)
The size of the matrix N is also a part of λ and the corresponding shift is −1.
5.1 Shape Invariance Data
Here are the shape invariance data of various polynomials [7].
1. Krawtchouk (2.41); λ = (p,N),δ = (0,−1), κ = 1, E(1;λ) = 1 ⇒ E(n;λ) = n.
23
2. Meixner (2.45); λ = (β, c), δ = (1, 0), κ = 1, E(1;λ) = 1 ⇒ E(n;λ) = n.
3. Charlier (2.48); λ = (a), δ = (0), κ = 1, E(1;λ) = 1 ⇒ E(n;λ) = n.
4. Racah (2.52); λ = (a, b, c, d), δ = (1, 1, 1, 1), κ = 1, E(1 ;λ) = 1 + d ⇒ E(n ;λ) =
n(n+ d).
5. Hahn (2.53); λ = (a, b,N), δ = (1, 1,−1), κ = 1, E(1;λ) = a + b ⇒ E(n;λ) =
n(n+ a+ b− 1).
6. dual Hahn (2.54); λ = (a, b,N), δ = (1, 0,−1), κ = 1, E(1;λ) = 1 ⇒ E(n;λ) = n.
7. q-Racah (2.55); qλ = (a, b, c, d), δ = (1, 1, 1, 1), κ = q−1, E(1 ;λ) = (q−1 − 1)(1 − dq)
⇒ E(n;λ) = (q−n − 1)(1− dqn).
8. q-Hahn (2.56); qλ = (a, b, qN), δ = (1, 1,−1), κ = q−1, E(1;λ) = (q−1 − 1)(1 − ab)
⇒ E(n;λ) = (q−n − 1)(1− abqn−1).
9. dual q-Hahn (2.57); qλ = (a, b, qN), δ = (1, 0,−1), κ = q−1, E(1 ;λ) = q−1 − 1
⇒ E(n ;λ) = q−n − 1.
10. quantum q-Krawtchouk (2.58); qλ = (p, qN), δ = (1,−1), κ = q, E(1 ;λ) = 1 − q
⇒ E(n ;λ) = 1− qn.
11. q-Krawtchouk (2.59); qλ = (p, qN), δ = (2,−1), κ = q−1, E(1 ;λ) = (q−1 − 1)(1 + pq)
⇒ E(n ;λ) = (q−n − 1)(1 + pqn).
12. dual q-Krawtchouk (2.60); qλ = (p, qN), δ = (1,−1), κ = q−1, E(1 ;λ) = q−1 − 1
⇒⇒ E(n ;λ) = q−n − 1.
13. affine q-Krawtchouk (2.61); qλ = (p, qN), δ = (1,−1), κ = q−1, E(1 ;λ) = q−1 − 1
⇒ E(n ;λ) = q−n − 1.
14. little q-Jacobi (2.62); qλ = (a, b), δ = (1, 1), κ = q−1, E(1 ;λ) = (q−1 − 1)(1 − abq2)
⇒ E(n ;λ) = (q−n − 1)(1− abqn+1).
15. little q-Laguerre (2.63); qλ = a, δ = 1, κ = q−1, E(1 ;λ) = q−1−1⇒ E(n ;λ) = q−n−1.
16. Al-Salam Carlitz II (2.64); qλ = a, δ = 0, κ = q, E(1 ;λ) = 1− q ⇒ E(n ;λ) = 1− qn.
24
17. alternative q-Charlier (2.65); qλ = a, δ = 2, κ = q−1, E(1 ;λ) = (q−1 − 1)(1 + aq)
⇒ E(n ;λ) = (q−n − 1)(1 + aqn).
For various q-polynomials, e.g. (dual) q-Hahn, q-Racah, etc. the shifts of the parameters
a, b, c, etc, are multiplicative. For example, in the q-Hahn case a → aq, b → bq. It should
be noted that some dual pairs, e.g. the q-Hahn and dual q-Hahn, q-Krawtchouk and dual
q-Krawtchouk, etc, have the identical parameter set λ but the shifts δ are different.
5.1.1 Exercise
Verify the above shape invariance (5.1) and the universal eigenvalue formula (5.2) for the
examples in §2.2.1, §2.2.2 with the data given above.
5.2 Auxiliary Function φ(x)
In the potential functions B(x;λ), D(x;λ) the parameter shift λ → λ+ δ is closely related
with the coordinate shift x → x + 1. But they are not the same for systems having the
sinusoidal coordinate η(x) = x. Let us introduce an auxiliary function φ(x;λ) by
φ(x;λ)def=η(x+ 1;λ)− η(x;λ)
η(1;λ), φ(0;λ) = 1, (5.6)
which connects potential functions B(x;λ + δ), D(x;λ + δ) with B(x + 1;λ), D(x;λ) in
systems with η(x) = x:
B(x;λ+ δ) = κ−1φ(x+ 1;λ)
φ(x;λ)B(x+ 1;λ), (5.7)
D(x;λ+ δ) = κ−1φ(x− 1;λ)
φ(x;λ)D(x;λ). (5.8)
Likewise the groundstate eigenvector ϕ0(x) with λ and λ+ δ are related√B(0 ;λ)ϕ0(x ;λ+ δ) = φ(x ;λ)
√B(x ;λ)ϕ0(x ;λ), (5.9)√
B(0 ;λ)ϕ0(x ;λ+ δ) = φ(x ;λ)√D(x+ 1 ;λ)ϕ0(x+ 1 ;λ). (5.10)
Crum’s theorem tells that the eigenvectors ϕn(x ;λ) and ϕn−1(x ;λ+ δ) are related by A(λ)
and A(λ)†:
A(λ)ϕn(x;λ) = E(n)(λ)ϕn−1
(x;λ+ δ
)/√B(0;λ), (5.11)
A(λ)†ϕn−1
(x;λ+ δ
)= ϕn(x;λ)×
√B(0;λ), (5.12)
in which the factor√B(0;λ) is introduced for convenience.
25
5.2.1 Exercise
1. Verify the relations among potential functions (5.7), (5.8) for various systems listed in
§2.2.1, §2.2.2.
2. Show that the relations satisfied by the groundstate vector ϕ0(x) (5.9), (5.10) are the
consequences of (5.7), (5.8) and (2.27).
5.3 Universal Rodrigues Formula
Let us introduce the forward and backward shift operators F(λ) and B(λ) acting on the
polynomial eigenfunctions {Pn(x ;λ)}:
F(λ)def= ϕ0(x;λ+ δ)−1 · A(λ) · ϕ0(x;λ)×
√B(0;λ)
= B(0;λ)φ(x;λ)−1(1− e∂), (5.13)
B(λ) def= ϕ0(x;λ)
−1 · A(λ)† · ϕ0(x;λ+ δ)× 1√B(0;λ)
=1
B(0;λ)
(B(x;λ)−D(x;λ)e−∂
)φ(x;λ), (5.14)
H = B(λ)F(λ) = B(x ;λ)(1− e∂) +D(x ;λ)(1− e−∂), (5.15)
F(λ)Pn(x ;λ) = E(n;λ)Pn−1(x ;λ+ δ), (5.16)
B(λ)Pn−1(x ;λ+ δ) = Pn(x ;λ). (5.17)
By taking the hermitian conjugate of the forward shift operator F(λ) (5.13), we obtain
A(λ)† ∝ ϕ0(x;λ)−1 · D(λ) · ϕ0(x;λ+ δ), (5.18)
D(λ)def= (1− e−∂)φ(x;λ)−1. (5.19)
By applying the above formula to the shape invariance formula (5.3) and imposing the
universal normalisation (2.38), we obtain universal Rodrigues formula for all the orthogonal
polynomials of a discrete variable in the Askey scheme:
Pn(η(x;λ);λ) = ϕ0(x;λ)−2 ·
n−1∏j=0
D(λ+ jδ) · ϕ20(x;λ+ nδ). (5.20)
This is especially simple for the systems with η(x) = x, Pn(x;λ) = ϕ0(x;λ)−2 · (1− e−∂)n ·
ϕ0(x;λ + nδ)2, which holds e.g. for the Krawtchouk (2.41), Meixner (2.45) and Charlier
(2.48). It is easy to convince that the universal normalisation Pn(0) = 1 is satisfied. In (5.20),
26
the term involving e−m∂ (1 ≤ m ≤ n) all vanish at x = 0 because of ϕ20(−m;λ + nδ) = 0
due to (2.51). The rest is simply ϕ0(0;λ)−2ϕ2
0(0;λ+nδ) = 1 due to the boundary condition
of ϕ0(x) at x = 0, (2.28).
5.3.1 Exercise
Calculate explicitly P1 and P2 of the Krawtchouk (2.41), Meixner (2.45) and Charlier (2.48)
by using the universal Rodrigues formula (5.20).
6 Solvability in Heisenberg Picture
In QM the Schrodinger picture and the Heisenberg picture are known. So far we have
discussed exactly solvable simplest QM in the Schrodinger picture. In this section we show
that the same systems are solvable in the Heisenberg picture, too.
In the Heisenberg picture, the observables depend on time and the states are fixed.
Heisenberg equation for the observable A(t) read
∂
∂tA(t) = i [H, A(t)]. (6.1)
By picking up an arbitrary normalised orthogonal functions {ϕn(x)}, one can rewrite the
above equation in matrix form:
∂
∂tAn,m(t) = i
∞∑k=0
(Hn,kAk,m(t)− An,k(t)Hk,m
),
An,m(t)def=
∫ϕn(x)
∗A(t)ϕm(x)dx, Hn,mdef=
∫ϕn(x)
∗Hϕm(x)dx.
The formal solution of the Heisenberg equation (6.1)
A(t) = eitHA(0)e−itH =∞∑n=0
(it)n
n!(adH)nA(0),
is expressed concisely in terms of adjoint operator (adH)Xdef= [H, X]. As seen from the
Heisenberg equation (6.1) (adH)nA(0) corresponds to the n-th time derivative of A(t) at
t = 0. For explicit closed form result one has to calculate multiple commutators
[H, [H, · · · , [H, A(0)]] · · · ]
and evaluate the infinite sum. This is why the Heisenberg equations are considered in-
tractable for most of the observables.
27
Here we show that the Heisenberg operator solution for the sinusoidal coordinate η(x)
eitHη(x)e−itH
can be obtained explicitly in a closed form for all the exactly solvable simplest QM [7,
17]. The sinusoidal coordinate η(x) is the argument of the eigenpolynomial Pn(η(x)) in the
factorised form of the eigenvector ϕn(x) (2.29). As shown in the preceding sections, for the
matrix interpretation of the Hamiltonian H, η(x) is a diagonal matrix. For the difference
operator interpretation, it is just a function.
6.1 Closure Relation
It is easy to see that the following closure relation is a sufficient condition for the exact
solvability of the Heisenberg equation for the sinusoidal coordinate,
[H, [H, η(x)] ] = η(x)R0(H) + [H, η(x)]R1(H) +R−1(H),
(adH)2η(x) = η(x)R0(H) + (adH)η(x)R1(H) +R−1(H),(6.2)
in which Ri(y), (i = 0,±1) is a polynomial in y. By similarity transformation in terms of
ϕ0(x ;λ) (2.27), (6.2) reads
[H, [H, η(x)] ] = η(x)R0(H) + [H, η(x)]R1(H) +R−1(H),
(ad H)2η(x) = η(x)R0(H) + (ad H)η(x)R1(H) +R−1(H),(6.3)
The l.h.s. consists of the operators e2∂, e∂, 1, e−∂, e−2∂ and H contains e±∂. Thus Ri can be
parametrised as
R1(z) = r(1)1 z + r
(0)1 , R0(z) = r
(2)0 z2 + r
(1)0 z + r
(0)0 , R−1(z) = r
(2)−1z
2 + r(1)−1z + r
(0)−1. (6.4)
The coefficients r(j)i depend on the overall normalisation of the Hamiltonian and the sinu-
soidal coordinate: r(j)1 ∝ B(0)1−j, r
(j)0 ∝ B(0)2−j, r
(j)−1 ∝ η(1)B(0)2−j.
Based on the closure relation (6.2), the triple commutator [H, [H, [H, η(x)]]] ≡ (adH)3η(x)
is expressed as a linear combination of η(x) and [H, η(x)] with polynomial coefficients in H,
(adH)3η(x) = [H, η(x)]R0(H) + [H, [H, η(x)]]R1(H)
= η(x)R0(H)R1(H) + (adH)η(x)(R1(H)2 +R0(H)
)+R−1(H)R1(H).
The closure relation (6.2) can be understood as the Cayley-Hamilton theorem for the operator
adH acting on η(x). The fact that the closure occurs at the second order reflects that the
28
Schrodinger equation is second order. Thus the higher order commutator (adH)nη(x) is also
expressed as a linear combination of η(x) and [H, η(x)] with polynomial coefficients in H.
We arrive at the desired result
eitHη(x)e−itH =∞∑n=0
(it)n
n!(adH)nη(x)
= [H, η(x)]eiα+(H)t − eiα−(H)t
α+(H)− α−(H)−R−1(H)R0(H)−1
+(η(x) +R−1(H)R0(H)−1
)−α−(H)eiα+(H)t + α+(H)eiα−(H)t
α+(H)− α−(H). (6.5)
This simply means that η(x) undergoes a sinusoidal motion with two frequencies α±(H)
depending on the energy:
α±(H)def= 1
2
(R1(H)±
√R1(H)2 + 4R0(H)
), (6.6)
α+(H) + α−(H) = R1(H), α+(H)α−(H) = −R0(H). (6.7)
The explanation is simple. The general solution of a three term recurrence relation with
constant coefficients
αan+2 + βan+1 + γan = 0, n = 0, 1, . . . , α, β, γ ∈ C,
is given by the roots of the corresponding quadratic equation,
αx2 + βx+ γ = 0, x±def=
−β ±√β2 − 4αγ
2α,
an = A(x+)n +B(x−)
n, A+B = a0, Ax+ +Bx− = a1.
The closure relation (6.2) becomes a three term recurrence relation by shifting η(x),
(adH)2η(x)− (adH)η(x)R1(H)− η(x)R0(H) = 0,
η(x)def= η(x) +R−1(H)R0(H)−1,
giving rise to the solution
(adH)nη(x) = Aα+(H)n +Bα−(H)n,
A = −η(x) α−(H)
α+(H)− α−(H)+ [H, η(x)] 1
α+(H)− α−(H),
B = η(x)α+(H)
α+(H)− α−(H)− [H, η(x)] 1
α+(H)− α−(H).
29
Summing them up produces eiα±(H)t in the solution (6.5).
The energy spectrum (eigenvalues) are determined by the overdetermined equations
(E(0) = 0)
E(n+ 1;λ)− E(n;λ) = α+(E(n;λ)) = E(1;λ+ nδ), (6.8)
E(n− 1;λ)− E(n;λ) = α−(E(n;λ)) = −E(1;λ+ (n− 1)δ). (6.9)
It is interesting to verify that the quantity R1(E(n))2 + 4R0(E(n)) inside of the square root
in α±(H) (6.6) is a complete square for exactly solvable systems.
6.2 Creation and Annihilation Operators
By combining the three term recurrence relation for Pn(η) (3.3) and the factorised form of
the eigenvectors (2.29), we obtain
η(x)ϕn(x) = Anϕn+1(x) +Bnϕn(x) + Cnϕn−1(x) (n ≥ 0). (6.10)
When the Heisenberg operator solution (6.5) is applied to the eigenvector ϕn(x), we obtain
eitHη(x)e−itHϕn(x)
= eit(E(n+1)−E(n))Anϕn+1(x) + Bnϕn(x) + eit(E(n−1)−E(n))Cnϕn−1(x)
= eitα+(E(n))Anϕn+1(x) +Bnϕn(x) + eitα−(E(n))Cnϕn−1(x) (6.11)
We find out creation a(+) and the annihilation a(−) operators as the coefficients of eiα±(H)t:
eitHη(x)e−itH
= a(+)eiα+(H)t + a(−)eiα−(H)t −R−1(H)R0(H)−1, (6.12)
a(±) def= ±
([H, η(x)]− η(x)α∓(H)
)(α+(H)− α−(H)
)−1
= ±(α+(H)− α−(H)
)−1([H, η(x)] + α±(H)η(x)
), (6.13)
a(+) † = a(−), (6.14)
a(+)ϕn(x) = Anϕn+1(x), a(−)ϕn(x) = Cnϕn−1(x), (6.15)
Bn = −R−1(E(n))R0(E(n))−1 = −(An + Cn). (6.16)
The above Bn formula (6.16) is quite interesting. It is independent of the normalisation
of the polynomials. The coefficients R0, R−1 are independent of the eigenpolynomial, as
30
they are determined by the Hamiltonian and the sinusoidal coordinate. But they are related
with the eigenpolynomial through the spectrum E(n). The eigenvectors of the excited states
{ϕn(x)} are obtained by multiple application of the creation operators on the groundstate
eigenvector ϕ0(x), ϕn(x) ∝ (a(+))nϕ0(x). This is the exact solvability in the Heisenberg
picture.
The eigenvector of the annihilation operator is the coherent state. It is interesting to
calculate the coherent states explicitly for various systems.
6.3 Determination of An and Cn
The starting point is the three term recurrence relations of the polynomial Pn(η) and its
dual Qx(E):
η(x)Pn(η(x)) = An(Pn+1(η(x))− Pn(η(x))
)+ Cn
(Pn−1(η(x))− Pn(η(x))
), (6.17)
E(n)Qx(E(n)) = B(x)(Qx(E(n))−Qx+1(E(n))
)+D(x)
(Qx(E(n))−Qx−1(E(n))
). (6.18)
By putting x = 0 in (6.18) and using Q0 = 1, D(0) = 0, we obtain
Pn(η(1)) = Q1(E(n)) =E(n)−B(0)
−B(0), (6.19)
in which the first equality is due to the duality (3.16). Next, by putting x = 1 in (6.17) and
using (6.19), we obtain
η(1)(E(n)−B(0)) = An(E(n+ 1)− E(n)) + Cn(E(n− 1)− E(n)). (6.20)
The two equations (6.16) and (6.20) for n ≥ 1 give
An =
R−1(E(n))R0(E(n))
(E(n)− E(n− 1)
)+ η(1)
(E(n)−B(0)
)E(n+ 1)− E(n− 1)
, (6.21)
Cn =
R−1(E(n))R0(E(n))
(E(n)− E(n+ 1)
)+ η(1)
(E(n)−B(0)
)E(n− 1)− E(n+ 1)
, (6.22)
and for n = 0 we obtain from (6.16)
A0 = R−1(0)R0(0)−1, C0 = 0. (6.23)
They are simplified (n ≥ 1) in terms of α±:
An =R−1(E(n)) + η(1)
(E(n)−B(0)
)α+(E(n))
α+(E(n))(α+(E(n))− α−(E(n))
) , (6.24)
31
Cn =R−1(E(n)) + η(1)
(E(n)−B(0)
)α−(E(n))
α−(E(n))(α−(E(n))− α+(E(n))
) . (6.25)
Note that (6.20) for n = 0 gives another important relation
A0E(1) +B(0)η(1) = 0. (6.26)
This relation ensures that the expression for Cn (6.25) vanishes for n = 0. Written differently,
the above relation (6.26) means another dual relation
B(0)
E(1)=
−A0
η(1), (6.27)
between the two intensive quantities. That is they are independent of the overall normali-
sation of the Hamiltonian H of the polynomial system.
6.4 Closure Relation Data and An, Cn
It is straightforward to calculate the closure relation (6.2), (6.3) for simpler systems. For
the Krawtchouk (2.41), Meixner (2.45) and Charlier (2.48), R1 = 0 and R0 = 1, R−1 = 0,
meaning α± = ±1. This leads to the linear spectrum E(n) = n (6.8), (6.9) and simple
expressions of the creation/annihilation operators (6.13), the coefficients of the three term
recurrence relation An, Cn (6.24), (6.25).
1. Krawtchouk (2.41); R1 = 0, R0 = 1, R−1(z) = (2p − 1)z − pN ⇒ An = −p(N − n),
Cn = −(1 − p)n. This polynomial is self-dual. That is, −An, −Cn are obtained from
B(x), D(x) by the replacement x → n. The dual Krawtchouk polynomial is exactly
the same as the Krawtchouk polynomial.
2. Meixner (2.45); R1 = 0, R0 = 1, R−1(z) = −1+c1−c z − βc
1−c ⇒ An = − c1−c(n + β),
Cn = − 11−c . The Meixner is also self-dual.
3. Charlier (2.48); R1 = 0, R0 = 1, R−1(z) = −z−a⇒ An = −a, Cn = −n. The Charlieris self-dual, too.
It is interesting to calculate the closure relation (6.2) and derive the coefficients of the three
term recurrence An, Cn explicitly (6.24), (6.25).
32
4. Racah (2.52); R1 = 2, R0(z) = 4z + d 2 − 1,
R−1(z) = 2z2 +(2(ab+ bc+ ca)− (1 + d)(1 + d)
)z + abc(d− 1),
⇒ −An=−(n+ a)(n+ b)(n+ c)(n+ d)
(2n+ d)(2n+ 1 + d), −Cn=−(n+ d− a)(n+ d− b)(n+ d− c)n
(2n− 1 + d)(2n+ d).
These are the same as B(x), D(x) of the Racah (2.52) with d replaced by d and x by
n.
5. Hahn (2.53); R1 = 2, R0(z) = 4z + (a+ b− 2)(a+ b),
R−1(z) = −(2N − a+ b)z − a(a+ b− 2)N ,
⇒ −An =(n+ a)(n+ a+ b− 1)(N − n)
(2n− 1 + a+ b)(2n+ a+ b), −Cn =
n(n+ b− 1)(n+ a+ b+N − 1)
(2n− 2 + a+ b)(2n− 1 + a+ b).
These are the same as B(x), D(x) of the dual Hahn (2.54) with x replaced by n.
6. dual Hahn (2.54); R1 = 0, R0 = 1, R−1(z) = 2z2 − (2N − a+ b)z − aN ,
⇒ −An = (n+ a)(N − n), −Cn = −n(b+N − n).
These are the same as B(x), D(x) of the Hahn (2.53) with x replaced by n.
7. q-Racah (2.55); R1(z) = (q−12 − q
12 )2z′, z′
def= z + (1 + d),
R0(z) = (q−12 − q
12 )2
(z′ 2 − (q−
12 + q
12 )2d
),
R−1(z) = (q−12 − q
12 )2
((1 + d)z′ 2 −
(a+ b+ c+ d+ d+ (ab+ bc+ ca)q−1
)z′
+((1− a)(1− b)(1− c)(1− dq−1)
+ (a+ b+ c− 1− dd+ (ab+ bc+ ca)q−1))(1 + d)
),
⇒ − An = − (1− aqn)(1− bqn)(1− cqn)(1− dqn)
(1− dq2n)(1− dq2n+1),
− Cn = −d (1− a−1dqn)(1− b−1dqn)(1− c−1dqn)(1− qn)
(1− dq2n−1)(1− dq2n).
These are the same as B(x), D(x) of the q-Racah (2.55) with d and d exchanged and
x replaced by n.
8. q-Hahn (2.56); R1(z) = (q−12 − q
12 )2z′, z′
def= z + (1 + abq−1),
R0(z) = (q−12 − q
12 )2
(z′ 2 − ab(1 + q−1)2
),
R−1(z) = (q−12 − q
12 )2
(z′ 2 −
(a(1 + bq−1) + (1 + aq−1)q−N
)z′
33
+ a(1 + q−1)((a− 1)bq−1 + (1 + bq−1)q−N
)),
⇒− An =(qn−N − 1)(1− aqn)(1− abqn−1)
(1− abq2n−1)(1− abq2n),
− Cn = aqn−N−1 (1− qn)(1− abqn+N−1)(1− bqn−1)
(1− abq2n−2)(1− abq2n−1).
These are the same as B(x), D(x) of the dual q-Hahn (2.57) with x replaced by n.
9. dual q-Hahn (2.57); R1(z) = (q−12 − q
12 )2z′, z′
def= z + 1, R0(z) = (q−
12 − q
12 )2z′ 2,
R−1(z) = (q−12 − q
12 )2
((1 + abq−1)z′ 2 −
(a(1 + bq−1) + (1 + aq−1)q−N
)z′
+ a(1 + q−1)q−N),
⇒ −An = (1− aqn)(qn−N − 1), −Cn = aq−1(1− qn)(qn−N − b).
These are the same as B(x), D(x) of the q-Hahn (2.56) with x replaced by n.
10. Quantum q-Krawtchouk (2.58); R1(z) = (q−12 − q
12 )2z′, z′
def= z − 1,
R0(z) = (q−12 − q
12 )2z′ 2,
R−1(z) = (q−12 − q
12 )2
(z′ 2 + p−1(1 + p+ q−N−1)z′ + p−1(1 + q−1)
),
⇒− An = p−1q−n−N−1(1− qN−n), −Cn = (q−n − 1)(1− p−1q−n).
11. q-Krawtchouk (2.59); R1(z) = (q−12 − q
12 )2z′, z′
def= z + 1− p,
R0(z) = (q−12 − q
12 )2
(z′ 2 + p(q−
12 + q
12 )2
),
R−1(z) = (q−12 − q
12 )2
(z′ 2 + (p− q−N)z′ + p(1 + q−1)(1− q−N)
),
⇒− An =(qn−N − 1)(1 + pqn)
(1 + pq2n)(1 + pq2n+1), −Cn = pq2n−N−1 (1− qn)(1 + pqn+N)
(1 + pq2n−1)(1 + pq2n).
12. affine q-Krawtchouk (2.61); R1(z) = (q−12 − q
12 )2z′, z′
def= z + 1,
R0(z) = (q−12 − q
12 )2z′ 2,
R−1(z) = (q−12 − q
12 )2
(z′ 2 − (pq + (1 + p)q−N)z′ + p(1 + q)q−N
),
⇒− An = (qn−N − 1)(1− pqn+1), −Cn = pqn−N(1− qn).
13. little q-Jacobi (2.62); R1(z) = (q−12 − q
12 )2z′, z′
def= z + 1 + abq,
R0(z) = (q−12 − q
12 )2
(z′ 2 − ab(1 + q)2
),
R−1(z) = (q−12 − q
12 )2
(−z′ 2 + (1 + a)z′ − a(1 + q)(1− bq)
),
⇒− An = aq2n+1 (1− bqn+1)(1− abqn+1)
(1− abq2n+1)(1− abq2n+2), −Cn =
(1− qn)(1− aqn)
(1− abq2n)(1− abq2n+1).
34
14. little q-Laguerre (2.63); R1(z) = (q−12 − q
12 )2z′, z′
def= z+1, R0(z) = (q−
12 − q
12 )2z′ 2,
R−1(z) = (q−12 − q
12 )2
(−z′ 2 + (1 + a)z′ − a(1 + q)
),
⇒− An = aq2n+1, −Cn = (1− qn)(1− aqn).
These are the same as B(x), D(x) of the Al-Salam Carlitz II (2.64) with x replaced by
n.
15. Al-Salam Carlitz II (2.64); R1(z) = (q−12−q 1
2 )2z′, z′def= z−1, R0(z) = (q−
12−q 1
2 )2z′ 2,
R−1(z) = (q−12 − q
12 )2
(z′ 2 + (1 + a)z′
),
⇒− An = aq−n, −Cn = (q−n − 1).
These are the same as B(x), D(x) of the little q-Laguerre (2.63) with x replaced by n.
16. alternative q-Charlier (2.65); R1(z) = (q−12 − q
12 )2z′, z′
def= z + 1− a,
R0(z) = (q−12 − q
12 )2
(z′ 2 + a(q−
12 + q
12 )2
),
R−1(z) = (q−12 − q
12 )2
(−z′ 2 + z′ − a(1 + q)
),
⇒− An = aq3n+1 1 + aqn
(1 + aq2n)(1 + aq2n+1), −Cn =
1− qn
(1 + aq2n−1)(1 + aq2n).
6.4.1 Exercise
Based on the above data of Rj(z), j = 0,±1, determine the eigenvalues {E(n)} as solutions
of the overdetermined equations (6.8), (6.9) for various systems.
7 Applications: Birth and Death Processes
Among many applications of Simplest Quantum Mechanics, let us discuss Birth and Death
Processes, [5]. They are typical stationary Markov processes with 1-dimensional discrete state
space, x = 0, 1, . . . , (N), . . . ,∞. That is the population of a group. Within an infinitesimal
time span, two things can happen to a group of population x, a birth with birth rate B(x) ≥ 0
takes x→ x+1 and a death with death rate D(x) ≥ 0 brings x→ x− 1, corresponding to a
tri-diagonal matrix. We show that all the exactly solvable simplest quantum mechanics give
exactly solvable birth and death processes [10, 14].
35
7.1 Schrodinger eqs. and Fokker-Planck eqs.
Let us review the well-known connection between the Fokker-Planck equation and the imag-
inary time Schrodinger equation [25]. To a 1-d Schrodinger equation i∂ψ(x;t)∂t
= Hψ(x; t)with a positive semi-definite hamiltonian H and a ground state eigenfunction ϕ0(x) > 0,
corresponds a 1-d Fokker-Planck equation
∂
∂tP(x; t) = LFPP(x; t), P(x; t) ≥ 0,
∫P(x; t)dx = 1, (7.1)
in which P(x; t) is the probability distribution. The Fokker-Planck operator LFP corre-
sponding to the Hamiltonian H is defined by [25]
LFPdef= −ϕ0 ◦ H ◦ ϕ−1
0 , (7.2)
This guarantees that the eigenvalues of LFP are negative semi-definite. The square nor-
malised groundstate eigenfunction ϕ0(x) provides the stationary distribution ϕ0(x)2 of the
corresponding Fokker-Planck operator:
∂
∂tϕ0(x)
2 = LFP ϕ0(x)2 = 0,
∫ϕ0(x)
2dx = 1. (7.3)
It is obvious that ϕ0(x)ϕn(x) is the eigenvector of the Fokker-Planck operator LFP :
LFPϕ0(x)ϕn(x) = −E(n)ϕ0(x)ϕn(x), n = 0, 1, . . . . (7.4)
Corresponding to an arbitrary initial probability distribution P(x; 0), (with∫P(x; 0)dx = 1),
which can be expressed as a linear combination of {ϕ0(x)ϕn(x)}, n = 0, 1, . . .,
P(x; 0) = ϕ0(x)∞∑n=0
cnϕn(x), c0 = 1, cndef= (ϕn, ϕ0(x)
−1P(x; 0)), n = 1, 2, . . . , (7.5)
we obtain the solution of the Fokker-Planck equation
P(x; t) = ϕ0(x)∞∑n=0
cn e−E(n)tϕn(x), t > 0. (7.6)
This is a consequence of the completeness of the eigenfunctions {ϕn(x)} (the polynomials)
of the Hamiltonian H. The positivity of the spectrum E(n) > 0, n ≥ 1 guarantees that the
stationary distribution ϕ20(x) is achieved at future infinity:
limt→∞
P(x; t) = ϕ20(x). (7.7)
36
The transition probability from y at t = 0 (i.e., P(x; 0) = δ(x− y)) to x at t is given by
P(x, y; t) = ϕ0(x)ϕ0(y)−1
∞∑n=0
e−E(n)tϕn(x)ϕn(y), t > 0. (7.8)
In terms of the polynomial Pn(η(x)), it is expressed as
P(x, y; t) = ϕ0(x)2
∞∑n=0
d2n e−E(n)tPn(η(x))Pn(η(y)), t > 0, (7.9)
in which dn is the normalisation constant (3.1).
7.2 Birth & Death Process
The birth and death equation is a discretisation of the Fokker-Planck equation in one di-
mension (7.1). It reads
∂
∂tP(x; t) = (LBDP)(x; t), P(x; t) ≥ 0,
∑x
P(x; t) = 1, (7.10)
with
LBDdef= −ϕ0 ◦ H ◦ ϕ−1
0 = (e−∂ − 1)B(x) + (e∂ − 1)D(x). (7.11)
The birth and death equation (7.10) in our notation reads
∂
∂tP(x; t) = −(B(x) +D(x))P(x; t) +B(x− 1)P(x− 1; t) +D(x+ 1)P(x+ 1; t). (7.12)
The transition probability from y at t = 0 (i.e., P(x; 0) = δx,y) to x at t has exactly the same
expression as that in the Fokker-Planck equation (7.8)
P(x, y; t) = ϕ0(x)ϕ0(y)−1
∑n=0
e−E(n)tϕn(x)ϕn(y), t > 0. (7.13)
In terms of the polynomial Pn(η(x)), or its dual Qx(E(n)), it is expressed as
P(x, y; t) = ϕ0(x)2∑n=0
d2n e−E(n)tPn(η(x))Pn(η(y)), t > 0, (7.14)
= ϕ0(x)2∑n=0
d2n e−E(n)tQx(E(n))Qy(E(n)), t > 0. (7.15)
It should be emphasised that in these formulas (7.13)-(7.14) everything is known includ-
ing the measure in contradistinction to the general formula by Karlin-McGregor [26]. The
transition probability satisfies the so-called Chapman-Kolmogorov equation [5]
P(x, y; t) =xmax∑z=0
P(x, z; t− t′)P(z, y; t′) (0 < t′ < t), (7.16)
as the consequence of the normalised eigenfunctionsxmax∑z=0
ϕn(z)ϕm(z) = δnm.
37
8 Summary and Comments
The eigenvalue problem of a certain type of hermitian matrices, i.e. tri-diagonal real sym-
metric (Jacobi) matrices, are explored as the simplest quantum mechanics. It is shown
that various concepts and solution techniques in exactly solvable QM in one dimension, e.g.
triangularity of the Hamiltonian with respect to certain polynomial flags spanned by sinu-
soidal coordinates, Crum’s theorem, shape invariance, Heisenberg operator solutions and
creation/annihilation operators, etc are also applicable in this simplest QM.
The present theory, known as discrete QM with real shifts [7, 8] provides unified under-
standing of various properties of the orthogonal polynomials of a discrete variable [13] in the
Askey scheme of hypergeometric orthogonal polynomials [4, 6]. The duality and dual polyno-
mials are emphasised as the characteristic features of these polynomials. They have various
applications in many different disciplines. Here we mention an interesting and important
one, the Birth and Death processes [10].
The orthogonal polynomials having Jackson integral measures [4] are not included, since
they require different settings [14]. The well known q-Meixner and q-Charlier are not covered,
too. They fall into the same category covered in this note and almost all the solution
methods, triangularity, shape invariance, Heisenberg operator solution, apply well. But the
eigenvectors do not form a complete set. Therefore the naive duality does not hold. See [14]
for comprehensive discussion on these excluded systems.
The present lecture is an elementary introduction to the theory of discrete QM with real
shifts. For a fuller comprehensive treatment, we refer to [7, 17] and [8]. The latter article is
a year 2011 review of discrete QM with real shifts as well as pure imaginary shifts. The well
known Askey-Wilson and Wilson systems are the typical constituents of the latter category.
In this introductory lecture, we have treated the exactly solvable potential functions
B(x), D(x) listed in §2.2.1, §2.2.2 as ‘given’. There is a unified theory to ‘construct’ these
potentials based on the properties of the sinusoidal coordinate η(x) [17].
Because of the time limitation, we are not able to cover a most exciting recent topic in
exactly solvable QM and orthogonal polynomials. That is, the construction of the exceptional
and multi-indexed orthogonal polynomials [18, 19, 20, 21, 22]. They form a complete set
of orthogonal polynomials but they do not satisfy three term recurrence relations, because
certain lowest degrees are missing. They are obtained as the main part of eigenfunctions
38
of exactly solvable quantum mechanical systems with second order differential/difference
Schrodinger equations, which are obtained by deforming known exactly solvable systems
through Darboux transformations [23].
Acknowledgements
R.S. thanks Pauchy Hwang, NTU, for the invitation to write this note.
A Symbols, Definitions & Formulas
◦ shifted factorial (Pochhammer symbol) (a)n :
(a)ndef=
n∏k=1
(a+ k − 1) = a(a+ 1) · · · (a+ n− 1) =Γ(a+ n)
Γ(a). (A.1)
◦ q-shifted factorial (q-Pochhammer symbol) (a ; q)n :
(a ; q)ndef=
n∏k=1
(1− aqk−1) = (1− a)(1− aq) · · · (1− aqn−1) =(a ; q)∞(aqn ; q)∞
. (A.2)
◦ hypergeometric function rFs :
rFs
(a1, · · · , arb1, · · · , bs
∣∣∣ z) def=
∞∑n=0
(a1, · · · , ar)n(b1, · · · , bs)n
zn
n!, (A.3)
where (a1, · · · , ar)ndef=
∏rj=1(aj)n = (a1)n · · · (ar)n.
◦ q-hypergeometric series (the basic hypergeometric series) rϕs :
rϕs
(a1, · · · , arb1, · · · , bs
∣∣∣ q ; z) def=
∞∑n=0
(a1, · · · , ar ; q)n(b1, · · · , bs ; q)n
(−1)(1+s−r)nq(1+s−r)n(n−1)/2 zn
(q ; q)n, (A.4)
where (a1, · · · , ar ; q)ndef=
∏rj=1(aj ; q)n = (a1 ; q)n · · · (ar ; q)n.
◦ differential equations
H, Hermite : ∂2xHn(x)− 2x∂xHn(x) + 2nHn(x) = 0, (A.5)
L, Laguerre : x∂2xL(α)n (x) + (α + 1− x)∂xL
(α)n (x) + nL(α)
n (x) = 0, (A.6)
J, Jacobi : (1− x2)∂2xP(α,β)n (x) +
(β − α− (α+ β + 2)x
)∂xP
(α,β)n (x)
+ n(n+ α + β + 1)P (α,β)n (x) = 0. (A.7)
39
◦ Rodrigues formulas
H : Hn(x) = (−1)nex2( d
dx
)ne−x
2
, (A.8)
L : L(α)n (x) =
1
n!
1
e−xxα
( d
dx
)n(e−xxn+α
), (A.9)
J : P (α,β)n (x) =
(−1)n
2nn!
1
(1− x)α(1 + x)β
( d
dx
)n((1− x)n+α(1 + x)n+β
). (A.10)
B Three Term Recurrence Relations
Any orthogonal polynomial starting from degree 0 constant satisfies three term recurrence
relations
ηQn(η) = AnQn+1(η) +BnQn(η) + CnQn−1(η), (B.1)
Q0(η) = constant, Q−1(η) = 0.
Here real constants An, Bn and Cn depend on the normalisation of the polynomial. Three
term recurrence relations are a simple consequence of the orthogonality.
((Qn, Qm))def=
∫W (η)Qn(η)Qm(η)dη = 0, n = m,
in whichW (η) is the orthogonality weight function. For its positive definiteness An−1Cn > 0
is necessary. Since ηQn(η) is a degree n+ 1 polynomial, it is expressed as
ηQn(η) = AnQn+1(η) +BnQn(η) + CnQn−1(η) +Dn(η),
in which Dn(η) is a degree k (0 ≤ k < n − 1) polynomial, if it is not vanishing. This
means ((Qk, Dn)) = 0 but this leads to a contradiction as ((Qk, ηQn)) = ((ηQk, Qn)) = 0 and
((Qk, Qn+1)) = ((Qk, Qn)) = ((Qk, Qn−1)) = 0 require Dn(η) ≡ 0. Conversely, any polyno-
mial starting from degree 0 and satisfying three term recurrence relations is an orthogonal
polynomial (Favard) [24].
References
[1] L. Infeld and T.E.Hull, “The factorization method,” Rev. Mod. Phys. 23 (1951) 21-68.
[2] See, for example, a review: F.Cooper, A.Khare and U. Sukhatme, “Supersymmetry
and quantum mechanics,” Phys. Rep. 251 (1995) 267-385.
[3] R. Sasaki, “Exactly solvable quantum mechanics,” The Universe 2 (2014) No.2 2-32;
arXiv:1411.2703[math-ph].
40
[4] G. E.Andrews, R.Askey and R.Roy, Special Functions , Encyclopedia of mathematics
and its applications, Cambridge, (1999).
[5] M.E.H. Ismail, Classical and quantum orthogonal polynomials in one variable, Ency-
clopedia of mathematics and its applications, Cambridge, (2005).
[6] R.Koekoek, P.A. Lesky and R.F. Swarttouw, Hypergeometric orthogonal polynomials
and their q-analogues, Springer-Verlag Berlin-Heidelberg (2010).
[7] S. Odake and R. Sasaki, “Orthogonal Polynomials from Hermitian Matrices,” J. Math.
Phys. 49 (2008) 053503 (43 pp), arXiv:0712.4106[math.CA].
[8] S. Odake and R. Sasaki, “Discrete quantum mechanics,” (Topical Review) J. Phys.
A44 (2011) 353001 (47 pp), arXiv:1104.0473[math-ph].
[9] S. Odake and R. Sasaki, “Exceptional (Xℓ) (q)-Racah polynomials,” Prog. Theor. Phys.
125 (2011) 851-870, arXiv:1102.0812[math-ph]; “Multi-indexed (q-)Racah polynomi-
als,” J. Phys. A45 (2012) 385201 (21 pp), arXiv:1203.5868[math-ph].
[10] R. Sasaki, “Exactly Solvable Birth and Death Processes,” J. Math. Phys. 50 (2009)
103509 (18 pp), arXiv:0903.3097[math-ph].
[11] M.M.Crum, “Associated Sturm-Liouville systems,” Quart. J. Math. Oxford Ser. (2) 6
(1955) 121-127, arXiv:physics/9908019.
[12] S.Odake and R. Sasaki, “Unified Theory of Annihilation-Creation Operators for Solv-
able (‘Discrete’) Quantum Mechanics,” J. Math. Phys. 47 (2006) 102102 (33pages),
arXiv:quant-ph/0605215; “Exact solution in the Heisenberg picture and annihilation-
creation operators,” Phys. Lett. B641 (2006) 112-117, arXiv:quant-ph/0605221.
[13] A.F.Nikiforov, S.K. Suslov, and V.B.Uvarov, Classical Orthogonal Polynomials of a
Discrete Variable, Springer, Berlin, (1991).
[14] S. Odake and R. Sasaki, “Orthogonal Polynomials from Hermitian Matrices II,”
arXiv:1604.00714[math.CA].
[15] D. Leonard, “Orthogonal polynomials, duality, and association schemes,” SIAM J.
Math. Anal. 13 (1982) 656–663.
[16] P.Terwilliger, “Leonard pairs and the q-Racah polynomials,” Linear Algebra Appl. 387
(2004) 235–276, arXiv:math.QA/0306301; “Two linear transformations each tridiagonal
with respect to an eigenbasis of the other; an algebraic approach to the Askey scheme
of orthogonal polynomials,” arXiv:math.QA/0408390.
41
[17] S. Odake and R. Sasaki, “Unified theory of exactly and quasi-exactly solvable ‘Discrete’
quantum mechanics:
I. Formalism,” J. Math. Phys. 51 (2010) 083502 (24 pp), arXiv:0903.2604[math-ph].
[18] D.Gomez-Ullate, N.Kamran and R.Milson, “An extension of Bochner’s problem: ex-
ceptional invariant subspaces,” J. Approx Theory 162 (2010) 987-1006, arXiv:0805.
3376[math-ph]; “An extended class of orthogonal polynomials defined by a Sturm-
Liouville problem,” J. Math. Anal. Appl. 359 (2009) 352-367, arXiv:0807.3939[math-
ph].
[19] D.Gomez-Ullate, N.Kamran and R.Milson, “Two-step Darboux transformations
and exceptional Laguerre polynomials,” J. Math. Anal. Appl. 387 (2012) 410-418,
arXiv:1103.5724[math-ph].
[20] C.Quesne, “Exceptional orthogonal polynomials, exactly solvable potentials and su-
persymmetry,” J. Phys. A41 (2008) 392001, arXiv:0807.4087[quant-ph]; B. Bagchi,
C.Quesne and R.Roychoudhury, “Isospectrality of conventional and new extended po-
tentials, second-order supersymmetry and role of PT symmetry,” Pramana J. Phys. 73
(2009) 337-347, arXiv:0812.1488[quant-ph].
[21] S.Odake and R. Sasaki, “Infinitely many shape invariant potentials and new orthogonal
polynomials,” Phys. Lett. B679 (2009) 414-417, arXiv:0906.0142[math-ph].
[22] S.Odake and R. Sasaki, “Exactly solvable quantum mechanics and infinite families of
multi-indexed orthogonal polynomials,” Phys. Lett. B702 (2011) 164-170, arXiv:1105.
0508[math-ph].
[23] G.Darboux, “Sur une proposition relative aux equations lineaires,” C. R. Acad. Paris
94 (1882) 1456-1459.
[24] T.Chihara, An Introduction to Orthogonal Polynomials, Gordon and Breach, (1978).
[25] H. Risken, The Fokker-Planck Equation (2nd. ed.), Springer-Verlag, Berlin, (1996).
[26] S.Karlin and J. L.McGregor, “The differential equations of birth-and-death processes,”
Trans. Amer. Math. Soc. 85 (1957) 489-546.
42
top related